1 Introduction

Eddy Covariance (EC) flux towers allow direct, high-frequency measurements of land-atmosphere carbon (C) exchange, providing a unique opportunity to evaluate C cycle processes in terrestrial ecosystems.

In grasslands — one of the most expansive and productive biomes covering over 40% of the Earth’s land surface yet highly heterogeneous — EC observations offer valuable insight into gross primary productivity (GPP) and its variability across climatic gradients.

However, global GPP estimates and underpinning EC flux datasets remain heavily weighted toward temperate regions, leaving tropical grasslands underrepresented and introducing uncertainty into global C estimates.

This study synthesised EC flux data at ~20 grassland sites predominantly in the tropical regions in Australia with publicly available grassland EC flux datasets including FLUXNET 2015, AMERIFLUX and ICOS, together with meteorological and remote sensing data to improve GPP estimation.

2 EC flux tower locations

2.1 Global grassland EC towers

  • Grassland map: WWF World Grassland Types 2014
  • Across publicly available EC flux datasets (FLUXNET 2015, AMERIFLUX and ICOS), the majority of grassland flux datasets are from the northern hemisphere and temperate regions (61/67 sites).
## Reading layer `WorldGrasslandTypes2014' from data source 
##   `C:\Users\takeda2\OneDrive - Queensland University of Technology\ec_flux_ml\WorldGrasslandTypes\WorldGrasslandTypes2014.shp' 
##   using driver `ESRI Shapefile'
## Simple feature collection with 145 features and 4 fields
## Geometry type: MULTIPOLYGON
## Dimension:     XY
## Bounding box:  xmin: -13150580 ymin: -5957072 xmax: 18405740 ymax: 7537678
## Projected CRS: World_Goode_Homolosine_Land

2.2 Australian grassland EC towers

  • Australian Bioregions
  • QUT has contributed 18/24 Grassland EC towers to those in Australia
## Reading layer `ibra7_regions' from data source 
##   `C:\Users\takeda2\daycent_r\au_bioregion\ibra7_regions.shp' 
##   using driver `ESRI Shapefile'
## Simple feature collection with 89 features and 11 fields
## Geometry type: MULTIPOLYGON
## Dimension:     XY
## Bounding box:  xmin: 72.57738 ymin: -54.77699 xmax: 167.9981 ymax: -9.14129
## Geodetic CRS:  GDA94

2.3 Latitudal distribution

3 Flux data overview

3.1 GPP

3.1.1 Date

3.1.2 Julian day

3.1.3 Temperature

3.1.4 VPD

## [1] "Sourced from meteorology data"

## [1] "Sourced from the flux tower"

3.1.5 Radiation

3.1.6 NDVI

3.1.7 Rain

## 
## Call:
## lm(formula = gpp ~ rain_rs15 * swrad_tower, data = df_flux_met_lsat_global_grass_filtered)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -18.4928  -1.9649  -0.8652   1.3499  21.8905 
## 
## Coefficients:
##                        Estimate Std. Error t value Pr(>|t|)    
## (Intercept)           6.301e-01  2.373e-02   26.55   <2e-16 ***
## rain_rs15             7.159e-03  5.364e-04   13.35   <2e-16 ***
## swrad_tower           6.445e-03  1.114e-04   57.88   <2e-16 ***
## rain_rs15:swrad_tower 1.708e-04  2.494e-06   68.46   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 3.225 on 165382 degrees of freedom
##   (924091 observations deleted due to missingness)
## Multiple R-squared:  0.251,  Adjusted R-squared:  0.251 
## F-statistic: 1.847e+04 on 3 and 165382 DF,  p-value: < 2.2e-16

4 GPP modelling

4.1 MODIS GPP model (MOD17A2)

## [1] "R-squared"
## [1] 0.5944789
## [1] "RMSE (gC/m2)"
## [1] 2.782233
## [1] "RRMSE (%)"
## [1] 93.20144
## [1] "PBIAS (%)"
## [1] -37.49346

4.2 Site-specific parameterisation

  • Using the 1st half of flux data at each site
    Parameter configuration
    Parameter Default value Lower limit Upper limit
    lue_max 0.86 0.5 3
    tmin_min -8.00 -25.0 0
    tmin_max 12.02 5.0 30
    vpd_min 650.00 50.0 1000
    vpd_max 5300.00 1500.0 6500

4.3 Site-specific parameter inference from MAP & MAT

4.4 Validation (daily)

  • Using the 2nd half of flux data at each site 
## [1] "R-squared"
## [1] 0.652315
## [1] "RMSE (gC/m2)"
## [1] 2.252509
## [1] "RRMSE (%)"
## [1] 74.74779
## [1] "PBIAS (%)"
## [1] -2.791108

4.5 Validation (annual)

## [1] "R-squared"
## [1] 0.7331842
## [1] "RMSE (gC/m2)"
## [1] 382.1422
## [1] "RRMSE (%)"
## [1] 47.99393
## [1] "PBIAS (%)"
## [1] -1.899277

5 GPP model application

6 Materials and Methods

7 Contact

For inquiries or collaborations, feel free to email me or connect via GitHub - the GitHub Issues page allows open discussion.

8 References

  • To be updated